This application pertains to improvements in linearization of radio frequency (RF) amplifiers to reduce the effects of intermodulation (IM) distortion.
All amplifiers are non-linear to some degree. If the signal carried by the amplifier has an envelope that fluctuates in magnitude, such as a multicarrier signal or a linear data modulation, then the non-linear operation generates intermodulation (IM) products in the amplifier output. These IM products represent unwanted interference in the operating band of the amplifier. Although it is possible to reduce the power of the IM products relative to the power of the desired signal by reducing the drive level of the amplifier, this expedient also reduces the power efficiency of the amplifier. Increasing the linearity of the amplifier by means of external circuitry can be a more efficient alternative.
A number of prior art approaches to this problem are described in U.S. Pat. No. 5,489,875, which is incorporated herein by reference. Some of the best prior art approaches described therein use a feedforward linearizer.
Traditional feedforward linearizers include a signal cancellation circuit and a distortion cancellation-circuit. The signal cancellation circuit has two branches, one of which contains the power amplifier whose output is to be linearized. In particular, the amplifier's output consists of an amplified version of an input signal, plus IM distortion. The other branch of the signal cancellation circuit contains circuitry characterized by a coefficient .alpha. (amplitude and phase) that can be adjusted to match the amplitude and phase shift of the amplifier, and a delay, also chosen to match the amplifier. If the match is perfect, the error signal obtained by subtracting the output of the two branches of the signal cancellation circuit equals the IM distortion. In the distortion cancellation circuit, an appropriately amplified and phase shifted version (coefficient .beta.) of the distortion is subtracted from the amplifier output, ideally leaving only the linearly amplified replica at the feedforward output.
FIG. 1 shows an example of a traditional prior art feedforward amplifier. The incoming signal is split by splitter S1 into two paths comprising the signal cancellation circuit. The first path 10, 15, 20 contains a complex gain adjuster CGA1 and the main amplifier A1, the output 20 of which contains the amplified desired signal and unwanted IM distortion. Splitter S2 directs part of the main amplifier output along line 25 to combiner C1. The second path 30, 35, 40 carries the desired signal, delayed by delay line DL1 to match the delay in the first path, to another input of combiner C1. The complex gain adjuster CGA1 provides means to change the amplitude and phase so that the signal component is cancelled in combiner C1, leaving only the IM distortion at line 45. The distortion cancellation circuit also consists of two branches. In one, the IM distortion on line 45 passes through complex gain adjuster CGA2 and auxiliary amplifier A2 to combiner C2, which receives at its other input 70 the main amplifier output, delayed in delay line DL2 to match the delay of path 25, 45, 50, 55, 60. When complex gain adjuster CGA2 is correctly adjusted, the IM distortion is cancelled in combiner C2, leaving only the amplified input signal at its output 75.
Typical implementations of the complex gain adjuster are shown for polar coordinates in FIG. 2(a) and for rectangular coordinates in FIG. 2(b). The input, output and two components of complex gain are denoted by I, O, GA and GB, respectively.
The complex gain adjuster CGA1 can alternatively be placed in line 30, although doing so precludes cancellation of any distortion introduced by the complex gain adjuster itself.
Because feed forward linearization is based on subtraction of nearly equal quantities, its major parameters must adapt to changes in operating environment, such as signal level, supply voltage and temperature.
The "minimum power" principle may be implemented in the prior art feedforward amplifier of FIG. 1. In the signal cancellation circuit, controller CT1 operates to minimize the power measured on line 100 using control lines 110 and 115 to complex gain adjuster CGA1. This approach does not make use of line 105. Instead, the system increments the voltages on control lines 110, 115 in the direction that results in a lower power measured on line 100.
The "gradient" method is an alternative to the minimum power principle for adaptation. FIG. 3 shows that the signal cancellation controller CT1 is a bandpass correlator. The signal for which the power is to be minimized at input I and a reference signal at input R are split in splitters S101, S102, respectively, and one of them is phase shifted by 90 degrees in phase shifter PS1. Two bandpass mixers M101, M102 produce outputs for which the mean value indicates the direction and size of increments to the complex gain components. Integrators I1, I2 remove high frequency noise and sum the increments to produce the complex gain components at outputs GA and GB. The controller therefore operates to bring the mean value of the gradient to zero. Numeric designations on the input and output lines indicate where the bandpass controller is connected in the signal cancellation circuit. Other embodiments of the gradient method adapt the control voltages to complex gain adjuster CGA1 similarly. The gradient method is faster than previously proposed minimum power methods and does not require deliberate misadjustments in order to determine the direction of change. However, it is sensitive to DC offset at the output of the mixers that create the gradient signal.
The gradient method can also be applied to adaptation of the FIG. 1 distortion cancellation circuit, as indicated in FIG. 3 by the numeric designations in parentheses. Specifically, controller CT2 operates to bring the mean value of the correlation between the signal on line 85 and the signal on line 95 to zero using control lines 120 and 125 to complex gain adjuster CGA2.
A number of more sophisticated approaches are also disclosed in the '875 patent. In one of these approaches, the delay, gain and phase differences are automatically adjusted according to a gradient principle, instead of merely adjusting the gain and the phase.
The '875 patent also discloses approximating the gradient as a sum of partial gradients taken over limited bandwidths. In the case of the distortion cancellation circuit, this allows calculation of the gradient over selected frequency bands that do not contain the amplified input signal, in order to reduce the masking effect. The use of limited bandwidth for each partial gradient allows use of digital signal processing technology to perform the calculation, thereby eliminating the DC offset that could otherwise cause convergence to an incorrect value.
The '875 patent also discloses automatically adjusting the differences to minimize the power at the output of the corresponding cancellation circuit. At each adjustment step, a set of measurements corresponding to perturbed values of the parameters (describing delay, gain and phase) is made. From these measurements, an estimate of the gradient of the power surface is formed. All the parameters describing delay, gain and phase are then adjusted in a direction opposite to the gradient, thereby effecting the greatest decrease in the power to be minimized. In the case of the distortion cancellation circuit, the power to be minimized is the sum of powers measured in selected frequency bands that do not contain the amplified input signal, in order to reduce the masking effect.
FIG. 4 depicts another prior art feed forward amplifier that is disclosed in the '875 patent. The input signal on line 5 enters the signal cancellation circuit, where splitter S1 produces two branches. The upper branch consists of the delay, gain and phase adjusting circuit DGPA1 (described below) between lines 10 and 15, the main amplifier A1, and line 25 to combiner C1. The lower branch consists of delay line DL1 and line 40 to combiner C1. The delay in delay line DL1 is selected to be approximately equal to the maximum delay expected in the main amplifier. When adjusted properly, the desired signal is cancelled on line 45, leaving only the distortion and noise generated in the upper branch. The distortion cancellation circuit also has two branches. The upper branch consists of delay line DL2 and line 70 to combiner C2. The lower branch consists of splitter S4, complex gain adjuster DGPA2, auxiliary amplifier A2 and line 60 to combiner C2. When adjusted properly, the distortion is cancelled on line 75, leaving only the desired signal. Controllers CT3 and CT4 operate to adapt the delay, gain and phase in the signal cancellation and the distortion cancellation circuits, respectively.
FIG. 5(a) shows the preferred embodiment of the delay, gain and phase adjustment circuits DGPA1, DGPA2 shown in FIG. 4. Line numbers and block numbers shown without parentheses are associated with connections in the signal cancellation circuit, whereas line numbers and block numbers shown in parentheses are associated with connections in the distortion cancellation circuit. This convention is followed consistently below. The signal is split in splitter S7 into a main branch consisting of splitter S8 and complex gain adjuster CGA3, and a delayed branch consisting of delay line DL3 and splitter S9. The branches are recombined in combiner C3. The delay in delay line DL3 is selected to be approximately equal to the difference between the maximum and minimum expected delays in the main amplifier over the range of operating conditions. Appropriate settings of complex gain adjusters CGA4 and CGA3 allow line 15 to carry an interpolation of the delayed signal on line 135 and the undelayed signal on line 150. Such interpolations can approximate the input signal with a delay ranging from zero to the delay of delay line DL3. Approximations of delays outside this range are also possible, but with decreasing accuracy.
FIG. 5(b) shows an alternative embodiment of delay, gain and phase adjustment circuits DGPA1, DGPA2. This embodiment forms on lines 170 and 205 the sum and the difference of the delayed and undelayed signals, respectively; combiner C5 being arranged to subtract its inputs 185 and 200. The circuit applies the complex gain adjustments to the sum and the difference before recombination. It has the advantage of reducing the degree of interaction between the two branches, so that complex gain adjuster CGA4 can be adjusted substantially independently of complex gain adjuster CGA3. Other linear combinations of the delayed and undelayed signals, and of additional signals at intermediate values of delay, are contemplated as falling within the scope of U.S. Pat. No. 5,489,875.
Although delay, gain and phase comprise three parameters, there are four control lines to the delay, gain and phase adjuster circuit (i.e. lines 102, 103, 107, 108 for delay, gain and phase adjuster DGPA1; and, lines 127, 128, 122, 123 for delay, gain and phase adjuster DGPA2). This allows an additional degree of freedom in compensating frequency dependent effects in the signal cancellation circuit, and the adaptation methods described below take full advantage of this degree of freedom.
The delay, gain and phase adjuster circuit can alternatively be placed in the lower branch of the signal cancellation circuit, on line 30, although doing so allows any distortions introduced in the delay, gain and phase adjuster circuit to appear at the final output 75 without being cancelled themselves.
The '875 patent also discloses making the delay, gain and phase adjuster circuit adaptive by the gradient principle. FIG. 6 shows a detailed view of controller CT3 (or CT4) in the signal cancellation circuit of FIG. 4. Again, line numbers and block numbers shown without parentheses are associated with connections in the signal cancellation circuit, whereas line numbers and block numbers in parentheses indicate connections in the distortion cancellation circuit. By means of dual bandpass correlators, each controlling one of the complex gain adjusters in FIG. 5(a) or in FIG. 5(b), the controller drives to zero the correlation between the undelayed input signal at input R and the distortion signal at input I and the correlation between the delayed input signal at input RD and the distortion signal at input I. The speed of convergence is determined by the gains of the several components in the adaptation loop.
Outputs GA, GB, GAD and GBD from the controller of FIG. 6 are connected to the corresponding inputs of the delay, gain and phase adjusters shown in FIGS. 5(a) and 5(b) through lines 107 (122), 108 (123), 102 (127) and 103 (128). Conversion from rectangular to polar coordinates decreases the convergence time if the complex gain adjusters CGA4, CGA3, CGA6 or CGA5 are implemented in polar coordinates, as shown in FIG. 2(a), but this is not essential.
The '875 patent also describes operating the controllers CT3 and CT4 according to a "partial gradient" principle, as illustrated in FIG. 7. A local oscillator LO1 shifts a selected narrow spectral region of the undelayed input signal at input R, the delayed input signal at input RD, and the fed back signal at line I to an intermediate frequency, where the bandpass correlations are performed substantially as in FIG. 6. Shifting and bandwidth limitation are performed in the mixer/bandpass filter combinations M2 and BPF2, M3 and BPF3, and M1 and BPF1. The bandwidth of the bandpass filters is significantly less than the operating bandwidth of the amplifier, so that only a partial gradient is produced. In operation, the output frequency of oscillator LO1 is stepped across the operating band. The sum of the resulting partial correlations is a good approximation to the full gradient calculated by the circuit in FIG. 6, and the integrators contained in bandpass correlators BPC3 and BPC4 inherently perform such a summation.
One advantage of the partial gradient principle applied to controller CT4 in the distortion cancellation circuit is that it can reduce the masking effect of the desired signal on line 85. In the case of a multicarrier signal, as shown in FIG. 8(a), the bandwidth of the bandpass filter is selected not to exceed the bandwidth of each carrier, and the output frequency of oscillator LO1 steps in increments of multiples of the minimum carrier spacing, selecting only those carrier locations that contain distortion and noise, with no component of the input signal. The resulting sum of selected partial correlations is only an approximation of the true gradient, but it has much improved signal to noise ratio. In the case of a single carrier, as shown in FIG. 8(b), the "skirts" of the spectrum contain IM distortion, and the partial gradients are calculated only in these skirts, and include no component of the desired signal. Similarly, but with less advantage, controller CT3 in the signal cancellation circuit can apply the partial gradient principle. In this case, the oscillator LO1 selects only those spectral regions that contain the carriers or the desired signal, in order to minimize masking effects of the distortion and noise.
A second advantage of the partial gradient controller is that the narrower bandwidth lends itself to implementation of the correlation operation by means of digital signal processing technology. As explained in the '875 patent, the DC offsets inherent in analog mixers can, in consequence, be eliminated. FIG. 9 shows one such prior art embodiment. As in FIG. 7, oscillator LO1 and mixer/bandpass filter combinations M2 and BPF2, M3 and BPF3, and M1 and BPF1 shift narrow spectral regions of the undelayed input signal at input R, the delayed input signal at input RD, and the fed back signal at input I all to an intermediate frequency f.sub.I. A second stage of down conversion shifts the outputs of bandpass filters to a range suitable for further processing in the digital signal processor DSP1. This second stage is accomplished by oscillator LO2, which produces a carrier at frequency f.sub.I -W/2, where W is the bandwidth of the bandpass filters BPF2, BPF3 and BPF1, and by mixer/bandpass filter combinations M5 and BPF5, M6 and BPF6, and M4 and BPF4. The outputs of the bandpass filters are centered at frequency W/2 and have bandwidth less than W. The DC offsets at the outputs of mixers M5, M6 and M4 are thereby eliminated. The bandpass filter outputs are then sampled at a rate at least equal to 2 W per second and converted to digital format in analog to digital converters ADC2, ADC3 and ADC1, from which they enter digital signal processor DSP1. The digital signal processor program operates as a pair of bandpass correlators to create the control signals GA, GB, GAD, and GBD through digital to analog converters DAC1, DAC2, DAC3 and DAC4.
If controllers CT3 or CT4 operating according to the partial gradient principle are employed with the sum and difference form of the delay, gain and phase adjuster circuit illustrated in FIG. 5(b), then the line designations 106 (121), 107 (102), 108 (123), 101 (126), 102 (127) and 103 (128) indicate the connections between the controller and the delay, gain and phase adjuster circuit.
A particular configuration of the partial gradient controller is applicable when delay variations in the main and auxiliary amplifiers are not significant, and it is sufficient to employ a complex gain adjuster instead of a full delay, gain and phase adjuster. FIG. 10 illustrates this prior art use of the partial gradient controller in the signal cancellation circuit, where the oscillator LO1 steps across the operating band, selecting frequency bands that contain the desired signal. Comparison with FIG. 7 demonstrates considerable simplification. When this simplified form of the partial gradient controller is used for controller CT4 of the distortion cancellation circuit, the oscillator selects frequency bands that contain IM products that do not contain the desired signal, in order to reduce the masking effect.
The digital signal processing implementation of the partial gradient controller is similarly simplified when delay variations in the main and auxiliary amplifiers are not significant, and it is sufficient to employ a complex gain adjuster instead of a full delay, gain and phase adjuster. FIG. 11 illustrates this prior art configuration; it is considerably simpler than FIG. 9, but it retains the advantage of eliminating DC offset. When this controller is employed in the signal cancellation circuit, as illustrated, the oscillator LO1 selects bands containing the desired signal. Conversely, when it is used for controller CT4 of the distortion cancellation circuit, the oscillator selects bands that contain IM products that do not contain the desired signal.